Useless text-coding idea #n (but a sit is cute, I shall ignore this and pretend it actually has some use).

Imagine a text, where each word is viewed as an integer. By, for example, splitting the text at whitespaces (this would leave punctuation as being parts of words, but this is not a critical problem). We can then convert each word to an octet vector (by, say, using UTF-8 encoding, since that seems so popular, these days). This octet vector, in turn, can be viewed as either a big-endian or little-endian 8n-bit unsigned integer.

Being an integer, it can be decomposed into its prime factors and these can them be emitted in some suitable order, using some simple framing protocol (using, say, 16- bit "prime ordinal", using 0 as a delimiter).

Obvsiouly, this restricts you from expressing words that happen to be prime, unless they're within the first 65535 primes and I haven't actually run any tests on this, to see how it seems to work out on actual test data. But other than being useless, I think it has cuteness potential.